LOCAS: multilabel mRNA localization with supervised contrastive learning | JoVE Visualize

Area of Science:

Computational biology
Molecular biology
Bioinformatics

Background:

Messenger RNA (mRNA) subcellular localization is vital for gene regulation and protein synthesis control.
Existing computational methods often use single-label classification, failing to address the multi-label nature of mRNA localization.
Current deep learning approaches struggle to model relationships between multiple mRNA localization sites.

Purpose of the Study:

To develop a novel computational framework, Localization with Supervised Contrastive Learning (LOCAS), for accurate multi-label mRNA subcellular localization prediction.
To enhance the modeling of relationships between multiple localization sites using supervised contrastive learning.
To improve the precision of spatial and temporal control of protein synthesis through better mRNA localization prediction.

Main Methods:

Integration of an RNA language model (RiNALMo) for high-quality sequence embeddings.
Application of supervised contrastive learning (SCL) to refine embedding spaces and ensure biologically meaningful clustering.
Introduction of an overlap-threshold-based similarity measure to handle overlapping localization labels.
Utilization of an ML-Decoder with a cross-attention mechanism for enhanced multi-label classification.

Main Results:

LOCAS achieved state-of-the-art performance on the RNALocate and RNALocate V2.0 benchmark datasets across all evaluation metrics.
Ablation studies confirmed the significant contributions of contrastive learning and the ML-decoder to improved multi-label classification accuracy.
The proposed method effectively models relationships between multiple mRNA localization sites.

Conclusions:

LOCAS provides a powerful and scalable solution for multi-label mRNA subcellular localization prediction.
Integrating RNA sequence representation learning with SCL significantly advances the field of mRNA localization prediction.
The findings pave the way for more precise understanding and control of gene expression through accurate mRNA localization.